cvs2svn: raw or cooked?

John Cavanaugh recommended that cvs2svn output a script consisting
of "cvs checkout" and "svn commit" commands.

Karl said that that is likely to be too expensive.

I had spent the morning looking at what's out there and thinking
through designs and writing down my thoughts (I'll mail those out in
the next message), and I'd just argued myself around from "do it all
in C" to "do it all in Perl and call cvs and svn" when I logged in to
check my mail.

Thanks to Greg Hudson for the NetBSD tip. I now have the TOC of the
NetBSD CVS repository. It's about a gigabyte uncompressed, so I
should have the whole thing by tomorrow morning. (-:

The NetBSD repository is probably not world's biggest, but it should
be big enough to be representative for performance testing.

I'm a little worried that the whole repository's metadata may not fit
in memory if cvs2svn naively tries to load it all in.

Right now I'm strongly thinking about implementing John's plan using
"cvs checkout" and "svn commit". At least as a prototype. If it's
doggy slow, then we'll know it. I don't really expect it to be doggy
slow, though.

-- 
                                        K<bob>
kbob_at_jogger-egg.com, http://www.jogger-egg.com/